Implementasi Faster R-CNN dengan Resnet-50 dalam Identifikasi Lesi pada Citra CT-SCAN

Abstract

Abnormalities of body tissues or a lesion can be brought on by different factors such as infections, autoimmune disease processes, metabolic problems, cancers and others. Lesion’s size tends to be minor and requires accuracy, knowledge, and high focus to be found, especially on CT-Scan images. The diagnosis of lesion identification will be significantly influenced by the subjectivity of experts, in this case doctor and other medical specialists. Therefore, a lesion detector is a necessity to help and also facilitate medical experts to reading CT-Scan images to find the results. In this study, in terms of identifying lesions contained in CT-Scan images, it was used with help by using Faster R-CNN with ResNet-50. This study used a total of 575 CT-Scan images, which went through some preprocessing like converting the form of 16 bit png images to the structure of Houndsfield Unit (HU), the use of filtering, which was followed by histogram equalization. The accuracy value obtained 90.51% with precision, recall, and f1 score sequentially of 100%, 90.51%, also 95.01%.